We Just Automated 75% of Your Client Requests

Market Context: Why This Matters Now

Vito opened by stepping back from the product to frame the broader industry shift underway. He acknowledged the common fear that DIY website builders represent an existential threat to agencies, but offered a grounding data point: the DIY market currently represents only about 5% of the overall professional services market. That figure, he argued, puts the threat in perspective.

However, he was careful not to dismiss the disruption entirely. His position is that AI is not simply enlarging the existing DIY segment — it is changing what DIY is capable of. AI can 02 03 04 05 06 07 08 09 10

now execute tasks that previously required professional skill, and this is causing services and software to converge. The barrier to entry is dropping again, just as it did when page builders like Elementor and Divi entered the market roughly twelve to fifteen years ago.

Those tools reduced the complexity of web development, created a whole new category of web professional, and brought many of the summit's attendees into the industry in the first place.

Vito's thesis is that we are now witnessing a second-order version of that same shift. A new layer is being wrapped around the agency business. The question is not whether the shift is real, but what agencies do in response to it.

The Core Problem: Creation vs. Delivery Time

The heart of Vito's argument rests on research Atarim conducted when first building the platform. They surveyed 587 web professionals with a single question: how long does it take you to build a five-page small business website on your own, with everything you need, just you and the browser?

The audience responses during the session ranged from one day to two days — answers broadly consistent with what the research found. The point Vito was building toward is the comparison between solo creation time and collaborative delivery time.

He illustrated the ratio with a simple analogy: building a single presentation slide in Canva or PowerPoint takes about five minutes. Delivering that same slide through a client approval process takes 30 minutes at minimum, and if both parties are iterating, potentially a full week. That 500% to 700% expansion in time between creation and delivery is the defining inefficiency of agency work.

The situation has grown more acute with AI. Where building a small business website might once have taken a day or two, AI-assisted development has compressed that to roughly three hours. But the delivery timeline — the back and forth with the client, the feedback cycles, the approvals — has not moved. It still takes four to six weeks. The ratio has therefore gotten worse, not better, as creation speed increases.

This is the problem Atarim was built to solve, and it is the frame through which all of the features demoed in this session should be understood.

The Philosophy Behind the Product: Digitising Human Experience

Vito was explicit about the conceptual foundation of Atarim. The product was never designed primarily around technology — it was designed around human behaviour. He described the original insight as observing how people communicate naturally, across centuries and millennia, and trying to digitise that experience.

Client feedback in the real world tends to be short, direct, and contextual. Phrases like "make this a little bigger," "change this text," or "make this smarter" are the natural

language of creative collaboration. They are three-to-five-word strings that make complete sense in context but become ambiguous without it.

The original version of Atarim captured this by letting clients click directly on a website element and leave a comment, attaching a screenshot, browser data, and element context automatically. This digitised the act of pointing at something and saying "that."

As the product evolved, and especially as AI was introduced in late 2025, the platform has moved beyond capturing that communication to actively improving and acting on it. The AI features in this session represent a progression from recording human feedback to interpreting it, clarifying it, reviewing the work independently, and now executing on it.

Feature 1: The Clarity AI — From Vague to Actionable

The first AI feature Vito demonstrated is what he called the Clarity system, named after an AI agent called "Claire" in the interface. The underlying problem it addresses is feedback vagueness.

Vito shared a specific data point: 67% of all creative communications sent via email are too vague to action in the first pass. They require a clarification cycle — going back to the client to ask "what do you mean by that?" — and that cycle typically costs a week in the delivery timeline.

Atarim's visual collaboration approach, where clients click on specific page elements to leave comments, reduced that 67% figure to 21%. Even with a screenshot and element data automatically attached, clients can still write vague comments. The Claire AI agent was built to close the remaining gap.

The analogy Vito used is Grammarly, which checks for grammatical errors in real time.

Claire checks for clarity in real time. When a client submits a comment that is flagged as vague, Claire activates automatically and, using the context of the element being commented on, generates a set of specific, actionable interpretations of what the client might mean.

In the live demo, a vague comment triggered Claire to suggest three specific options:

• Make the header bigger and bolder
• Increase the spacing above and below the heading
• Darken the heading colour for better readability Each suggestion is selectable by the client with a single click, replacing the vague original text with a precise instruction. Vito also noted that a deeper explanatory layer exists beneath each suggestion for clients who want more detail.

The target is to bring the vague communication rate from 21% down to zero. Vito was careful to characterise this as mimicking what a good project manager or agency owner already does: going back to the client and asking "do you mean this or that?" — but doing it instantly and automatically.

At the time of the session, Atarim had also added a "Show Me" component to this part of the interface, which previews the suggested change visually before any decision is made.

Feature 2: AI Review System with Specialist Agents

The second major feature Vito demoed is Atarim's AI Review system. This addresses a different phase of the delivery problem: the internal review process.

Vito identified between five and eight distinct review sessions that should happen during a typical project:

• When a Figma design is handed off to a developer for implementation
• When client copy is placed onto the page and needs to be checked against the original document
• Before sharing work with a client (catching missing placeholder text, disconnected buttons, non-functional forms)
• Pre-launch review
• Post-launch review Each of these sessions takes 45 to 60 minutes when done by a human. In a standard project, that adds up to anywhere from four to eight hours of review time across five or more sessions. If the agency owner is doing this personally, which Vito acknowledged is common, it becomes a significant drain on their time. If a dedicated QA person is used, the cost scales.

The AI Review system runs these sessions automatically. It operates through a set of specialist agents, each focused on a specific area:

• Pixel — handles design review and UI concerns
• Lexi — handles content review
• Clero — the orchestrating agent that coordinates the other agents and synthesises outputs
• Additional agents covering SEO, functionality, and accessibility (which can be toggled on or off depending on the type of review) Crucially, Atarim's review operates at the element level rather than the page level. This is the same technical foundation as the visual collaboration tool: the system can inspect individual HTML elements, their properties, their spacing, their content, and their relationships. Vito contrasted this with how general-purpose LLMs like Claude Code approach page review — from a zoomed-out, external perspective that lacks element-level fidelity.

Atarim benchmarked this claim formally. They set up an AI judge, created evaluation criteria designed to remove bias, and compared three outputs for the same website: human QA review, Claude Code review, and Atarim AI review. Atarim's result scored approximately twice as well as the human review benchmark on the criteria tested. Vito acknowledged the obvious question — who is better, a cheap human reviewer or an expensive agency owner — and noted that having the agency owner do all QA is itself a misuse of time.

Canned Review Prompts One of the more practically useful elements of the review demo was the canned prompts system. Users can save custom prompts as reusable templates, standardising the review criteria for different project types or client categories.

Vito showed two examples from real users: A "compliance review" prompt specifying that a Figma file and a copy document had been provided, asking the AI to perform a strict design and copy compliance audit and defining specific rules — such as how to handle line breaks — that the reviewer should apply.

A "pre-client review" prompt designed for a small business website on a tight budget.

This prompt explicitly told the AI to skip advanced suggestions, focus on quick wins, avoid suggesting fundamental changes (because it was too late in the project for those), and prioritise only high-impact issues that would cause embarrassment in front of the client.

These examples illustrate how the review system can be tailored to context rather than running a generic audit.

Review Output After running a review, the system produces:

• A strategic overview of gaps (e.g., "the value proposition is slightly vague," "the CTA intent is diluted")
• A definition of what "great" looks like for that page
• An estimated lift if the recommendations are implemented
• An estimate of implementation time, calibrated to the user's experience level (defined in the user profile)
• A task list of specific, actionable issues
• Individual comments pinned to the relevant page elements In the demo, Pixel flagged that a form input container lacked sufficient breathing room around it. Vito clicked "Show Me" and the AI rendered a preview of the suggested change on screen, non-destructively.

Feature 3: "Show Me" and "Do It" — Closing the Loop

The "Show Me" and "Do It" features represent the newest layer of the platform and the most direct mechanism behind the 75% automation claim.

Show Me is a non-destructive preview tool. When the AI (or a client, or the agency owner) suggests a change to a page element, clicking "Show Me" renders that change visually in the browser without writing anything to the database or the underlying theme. A page refresh returns everything to the original state. This allows all parties to evaluate a proposed change before committing to it.

Do It executes the change. For WordPress sites, this connects directly to the Atarim WordPress plugin and applies the modification to the live (or staging) environment. An undo button is available, though Vito noted that if the element has been subsequently modified directly inside the page builder, the undo function may not be able to reverse the change cleanly.

Both features were described as being in beta at the time of the session. The "Show Me"

functionality was pushed live "literally hours ago" before the demo.

The 75% Figure The 75% automation statistic is derived from a specific evaluation. Atarim took the last 8,000 tasks created inside the platform and ran them through an AI evaluation system to determine what percentage of them could be resolved by the "Do It" button. The result was 75%.

Vito framed the nature of these tasks: they are micro-requests from clients. Things like "can you change this text to this," "can you move this a little to the left," "can you move this up or down." These are exactly the kinds of small, repetitive, time-consuming tasks that eat agency hours without requiring strategic thought. The 75% figure means that three-quarters of these requests can now be resolved without a developer touching the code manually.

Feature 4: The Project Brief and AI Context

Woven through the demo was a feature that underpins the quality of all the AI outputs: the one-click Project Brief.

Vito demonstrated that before his session, he had clicked a single button to generate a full project brief for the Atarim website. Within one minute, the system had produced a structured document covering:

• Brand assets
• Inspiration websites and visual references
• Tone of voice and style preferences
• Design preferences
• A description of the product and services offered
• Business overview and unique value proposition
• The primary goal of the project Vito noted he made zero edits to the generated brief. This document is fed as context into every AI interaction on that project — reviews, content suggestions, tag line generation, and clarification responses — so that all AI outputs are grounded in the specific client's business and brand, not generic best practices.

He demonstrated this directly. After the review generated a comment suggesting the homepage tagline could be improved, Vito asked the AI in real time for a better tagline. The AI returned "Get senior-level feedback on every page in seconds" and explained why: it preserved the core promise of the original text but replaced the passive framing with a more direct, benefit-led statement that spoke to speed and quality — the two things the brief had identified as central to Atarim's value proposition.

Vito clicked "Show Me" on the new tagline and then "Do It," and the change rendered on screen. He highlighted the undo button as a safety net.

The "From Feedback to Done" Vision

Throughout the session, Vito returned to a phrase that has become a central message for the platform: "From Feedback to Done." He described this as a tagline or notice appearing across the Atarim interface, summarising the direction the product has moved.

The arc of the platform's development, as described in this session, follows a clear progression:

1. Capture feedback better (visual collaboration, click-to-comment)
2. Improve feedback quality in real time (Claire / Clarity AI)
3. Automate internal review (AI Review with specialist agents)
4. Generate a brief to contextualise all AI work (one-click Project Brief)
5. Preview suggested changes (Show Me)
6. Execute suggested changes automatically (Do It) Together, these features are intended to collapse the four-to-six-week delivery timeline by reducing the friction at every stage where human back-and-forth currently introduces delay.

What is Coming Next

Vito teased a forward-looking session planned for Day 4 of the summit. He indicated he would be covering his predictions for the next two years of the web agency industry, drawing on his perspective as a founder who has gone from freelancer to agency owner to active industry participant over the past six years.

He described it as containing "a lot of hot takes" and said it would go beyond the Atarim product itself to look at the broader trajectory of the agency business in an AI-driven market.

Andrew Palmer, who co-hosted the intro and outro of this session, confirmed there would be two sessions on Day 4 touching on the future of the agency and the AI agency of the future.

About the speaker Vito Peleg Founder and CEO of Atarim

Vito Peleg is founder and CEO of Atarim, a visual collaboration and website feedback platform used by agencies and freelancers. He builds and speaks on AI tooling for real agency workflows.